Agents
Agents are reusable AI workers that combine instructions, knowledge, context, and approved tools.
An agent is the unit you publish, embed, or call from the API. It defines how Fluo should respond to a task and which resources are available during execution.
What An Agent Contains
| Part | Description |
|---|---|
| Instructions | The agent's role, tone, boundaries, and task rules. |
| Output behavior | Whether the response should be conversational, concise, structured, or formatted for another system. |
| Knowledge | Documents and sources the agent can use for grounded answers. |
| Context | Structured state or memory supplied during a run or thread. |
| Integrations | Approved external capabilities the agent may use. |
Release Surfaces
The same agent can support multiple release paths:
- A hosted chatbot embedded on a website.
- A one-off API run from your backend.
- A streaming thread in a custom product UI.
- An internal workflow that executes on demand.
Design Tips
- Give each agent a narrow job.
- Write instructions for real user scenarios.
- Add explicit refusal and escalation rules.
- Test with examples that should fail, not only examples that should pass.
- Keep public-facing agents away from internal knowledge unless it is intended for users.
Public vs Internal Agents
Public agents should be stricter than internal agents. Before publishing one, review:
- Which knowledge bases it can access.
- Which integrations it can call.
- Whether tools can write to external systems.
- What metadata your application sends with requests.
- Which domains and keys can reach it.